GNU coreutils 9.9: Detailed Test report 12/19

[ChrisDryden]

GNU coreutils Test Coverage Tracking

Failing GNU tests as of 2025-12-19 | Test Coverage Dashboard

Overview

With the number of failing tests now small enough to track individually, this document provides detailed descriptions of each remaining test failure and the next steps needed to achieve full GNU compatibility. Theres a bunch of duplicate PR's for the same tests in the PR queue now and I'm hoping this will allow us to consolidate our efforts.

I am also hoping to separate the tests that are incompatible because they either intercept glibc calls or call gdb on specific lines of source code since we are unable to fix those tests and it wouldn't be good to track compatibility of those tests.

---

FAILED

[ ] cp/cp-a-selinux.sh - PR #10207

GNU Test | Fix cp -a SELinux xattr handling and -z flag for folders

[x] cp/link-heap.sh - PR #9805 ✓

GNU Test | Reduce memory usage for cp -al by skipping unnecessary tracking - MERGED

[x] cp/preserve-gid.sh - PR #9803 ✓

GNU Test | Fix preserve-gid when canonicalize fails due to inaccessible parent dirs - MERGED

[ ] date/date-debug.sh

GNU Test | Debug output for date parsing

[ ] date/date-ethiopia.sh

GNU Test | Ethiopian calendar translation support

Note: PR #9661 was an attempt but has been closed.

[ ] date/date-iran.sh

GNU Test | Iranian/Persian calendar translation support

Note: PR #9661 was an attempt but has been closed.

[x] date/date-locale-hour.sh - PR #9654 ✓

GNU Test | Locale-aware 12h vs 24h format detection via nl_langinfo() - MERGED

[ ] date/date-thailand.sh

GNU Test | Thai calendar translation support

Note: PR #9661 was an attempt but has been closed.

[ ] date/date.pl

GNU Test | Comprehensive date parsing tests

[x] dd/no-allocate.sh - PR #9806 ✓

GNU Test | Use ibs/obs-sized buffer for skip/seek on non-seekable files (ulimit -v tests) - MERGED

[x] dd/nocache_eof.sh - PR #9818 ✓

GNU Test | Fix nocache flag handling at EOF (use len=0 for posix_fadvise) - MERGED

[x] dd/nocache_fail.sh - PR #9820 ✓

GNU Test | Use actual filename in nocache error messages - MERGED

[x] dd/skip-seek-past-file.sh - PR #9821 ✓

GNU Test | Use seek for stdin skip when possible - MERGED

[ ] dd/stderr.sh - PR #9664

GNU Test | Detect closed stdin before Rust sanitizes to /dev/null

[x] env/env-signal-handler.sh - PR #9465 ✓

GNU Test | --default-signal, --block-signal, --list-signal-handling flags - MERGED

[x] fmt/non-space.sh - PR #9329 ✓

GNU Test | Handle invalid UTF-8 input by replacing malformed sequences - MERGED

[ ] fold/fold-zero-width.sh - PR #9274

GNU Test | Zero-width character handling in fold

[x] ls/ls-misc.pl - PR #9262 ✓

GNU Test | Align symlink coloring with GNU semantics (or=, mi=, ln=target) - MERGED

[ ] ls/stat-free-symlinks.sh - PR #8728

GNU Test | Iterative loop for ls recursive directory search (prevents stack overflow)

[ ] misc/close-stdout.sh

GNU Test | Proper handling of closed stdout

[ ] misc/stdbuf.sh

GNU Test | Line/block buffering with fifos

[x] misc/tsort.pl - PR #9289 ✓

GNU Test | Topological sort tests - MERGED

[ ] misc/unexpand.pl - PR #9265

GNU Test | Tab unexpansion edge cases

[ ] misc/write-errors.sh

GNU Test | Write error handling and reporting

[x] mv/i-3.sh - PR #9599 ✓

GNU Test | TTY prompting for non-writable destination files - MERGED

[ ] numfmt/numfmt.pl

GNU Test | Number formatting comprehensive tests

[ ] numfmt/mb-non-utf8.sh - PR #10079

GNU Test | Support non-UTF8 delimiters for locales like GB18030 (test enabled by PR #9807 ✓)

[ ] ptx/ptx-overrun.sh - PR #9816, PR #9823

GNU Test | Fix overflow panic with long references + handle duplicate input files

[x] ptx/ptx.pl - PR #9681, PR #9682 ✓

GNU Test | Permuted index generation tests - MERGED (column width/padding + -S flag)

[ ] rm/one-file-system.sh - PR #7569

GNU Test | --one-file-system and --preserve-root=all implementation

[ ] rm/rm1.sh - PR #9011

GNU Test | Fix error reporting for rm -r on Linux

[x] runcon/runcon-no-reorder.sh - PR #9607 ✓

GNU Test | SELinux context argument ordering - MERGED (was passing incorrectly due to false symlinks)

[x] runcon/runcon-compute.sh - PR #10088 ✓

GNU Test | Fix PATH handling for runcon -c (don't search PATH for relative commands) - MERGED

[x] seq/seq-epipe.sh - PR #9657 ✓

GNU Test | SIGPIPE handling macro for proper exit codes - MERGED

[x] shred/shred-passes.sh - PR #9317 ✓

GNU Test | Deterministic pass sequence + stop on write errors - MERGED

[x] sort/sort-continue.sh - PR #9107 ✓

GNU Test | Respect soft fd limit for merge batch sizes - MERGED

[x] sort/sort-debug-keys.sh - PR #9468 ✓

GNU Test | Debug key annotation alignment - MERGED

[x] sort/sort-debug-warn.sh - PR #9900 ✓

GNU Test | Warning messages for obsolescent keys (+field -field) and options - MERGED

[ ] sort/sort-float.sh - PR #9839

GNU Test | Floating point number sorting

[ ] sort/sort-h-thousands-sep.sh - PR #9848

GNU Test | Human-readable sort with thousands separators

[ ] sort/sort-merge-fdlimit.sh - PR #9849

GNU Test | Merge sort file descriptor limits

[ ] sort/sort-month.sh - PR #9722

GNU Test | Locale-aware month parsing (ICU), handles "janv." and Latin-1

[ ] sort/sort.pl - PR #9862

GNU Test | GNU coreutils compatibility for comprehensive sort tests

[x] stty/stty-row-col.sh - PR #9516 ✓

GNU Test | Hex/octal parsing for rows and columns - MERGED

[x] stty/stty.sh - PR #9490 ✓

GNU Test | COLUMNS env support - MERGED

[ ] tac/tac-2-non-seekable.sh - PR #9664

GNU Test | Detect closed stdin before Rust sanitizes to /dev/null

[ ] tail/follow-name.sh

GNU Test | Follow renamed/recreated files

[ ] tail/follow-stdin.sh - PR #9664

GNU Test | Detect closed stdin before Rust sanitizes to /dev/null

[ ] tail/inotify-rotate-resources.sh

GNU Test | Inotify resource handling during log rotation

[x] tail/pid-pipe.sh - PR #9663 ✓

GNU Test | --pid with FIFO using non-blocking open - MERGED

[x] tail/pipe-f.sh - PR #10156 ✓

GNU Test | Detect broken stdout pipe and don't follow stdin - MERGED

[ ] tail/symlink.sh - PR #10158

GNU Test | Symlink following behavior

[x] tail/tail.pl - PR #10155 ✓

GNU Test | Fix big number handling (saturating_mul) and error message quoting - MERGED

[x] wc/wc-cpu.sh - PR #9144 ✓

GNU Test | Respect GLIBC_TUNABLES to disable AVX/AVX512 SIMD - MERGED

SKIPPED

[x] cp/cp-mv-enotsup-xattr.sh - PR #9530 ✓

GNU Test | xattr ENOTSUP handling - TEST ENABLED

[x] df/skip-rootfs.sh - PR #10159 ✓

GNU Test | Add rootfs to is_dummy_filesystem - MERGED

[ ] id/gnu-zero-uids.sh

GNU Test | Handling of UID 0

[x] id/smack.sh - PR #9910 ✓

GNU Test | SMACK security labels - adds --context flag and SMACK label display - MERGED

[x] mkdir/smack-no-root.sh - PR #9910 ✓

GNU Test | SMACK labels without root - adds --context flag for setting SMACK labels - MERGED

[x] mkdir/smack-root.sh - PR #9910 ✓

GNU Test | SMACK labels with root - adds --context flag for setting SMACK labels - MERGED

[x] mkdir/writable-under-readonly.sh - PR #9530 ✓

GNU Test | Creating writable dir under readonly parent - TEST ENABLED

[ ] pr/pr-tests.pl

GNU Test | Page formatting tests

[x] rm/fail-eperm.xpl - PR #9525, PR #9531 ✓

GNU Test | EPERM error on files not owned by user - MERGED/FIXED

[x] tac/tac-continue.sh - PR #10094 ✓

GNU Test | Use temp file for stdin (like GNU) instead of buffering in RAM; detect write errors and continue to next file - MERGED

[ ] tail/inotify-dir-recreate.sh

GNU Test | Directory recreation with inotify

[ ] timeout/timeout-group.sh

GNU Test | sigtimedwait() instead of 100ms polling (100ms → 1ms)

INCOMPATIBLE

Tests that use LD_PRELOAD to intercept glibc calls or gdb breakpoints. If we don't use the same internal calls, we will have to patch these tests.

cp/nfs-removal-race.sh (LD_PRELOAD) - PR #10086

GNU Test | Use FileInformation without dereference for symlink destination check

GNU test updated (commit 1afe410) to support both glibc < 2.33 and >= 2.33. PR uses FileInformation helpers to match how GNU fetches file metadata.

[x] csplit/csplit-io-err.sh - PR #9855 ✓

GNU Test | I/O error handling

PR #9855 detects and reports write errors - MERGED. GNU test updated (commit 1b46778) to use /dev/full instead of LD_PRELOAD. No GNU patch needed after rebase.

df/no-mtab-status.sh (LD_PRELOAD) - PR #9759 (test enabled)

GNU Test | Intercepts glibc to simulate missing mtab

PR #9759 enabled this test (detects #9760).

df/skip-duplicates.sh (LD_PRELOAD)

GNU Test | Intercepts glibc to inject duplicate mount entries

nproc/nproc-quota.sh (LD_PRELOAD)

GNU Test | Intercepts glibc to simulate CPU quota

rm/r-root.sh (LD_PRELOAD + gdb)

GNU Test | Uses gdb breakpoints and LD_PRELOAD for safety net

rm/rm-readdir-fail.sh (LD_PRELOAD) - PR #9604

GNU Test | Intercepts readdir to simulate failures

PR #9604 wraps libc calls to handle LD_PRELOAD tests.

stty/bad-speed.sh (LD_PRELOAD) - PR #9517

GNU Test | Intercepts cfsetispeed to simulate failures

PR #9517 implements input/output baud rate setting so cfsetispeed is called.

tail/inotify-race.sh (gdb) - PR #9908

GNU Test | Uses gdb breakpoints to trigger race conditions

PR #9908 patches the test to use Rust gdb breakpoints (breaking on watcher_rx.watch_with_parent in watch.rs).

tail/inotify-race2.sh (gdb) - PR #9908

GNU Test | Uses gdb breakpoints to trigger race conditions

PR #9908 patches the test to use Rust gdb breakpoints (breaking on watcher_rx.watch_with_parent in watch.rs).

ERROR

[ ] misc/usage_vs_getopt.sh - PR #9869

GNU Test | Usage message consistency with getopt

The coreutils multicall binary doesn't use clap for argument parsing, so unrecognized options like --unknown were outputting "function/utility not found" instead of "unrecognized option". PR #9869 fixes this by detecting option-like arguments.

[ChrisDryden]

I tried to create a script that would match, if the test name appeared in the comments and it appears that it did not work correctly, I am going through each of them and manually searching.

[sylvestre]

dup of #9127 ?

[ChrisDryden]

It is to some extent, I am unable to modify it, I've been adding comments to it for the last few weeks and I was hoping to have a more in depth description on each of the tests. Its pretty out of date at this point in the main one

[sylvestre]

ok, but why not asking before creating yet a new issue ? :)

[sylvestre]

i think i gave you permissions to edit issues

[ChrisDryden]

Thanks! I really appreciate it

[ChrisDryden]

I don't think the triage role has the ability to edit issues, I was doing some investigating to see if that was something that can be configured, but it seems this is a common issue that comes up: https://github.com/orgs/community/discussions/5773

[naoNao89]

tests/basenc/bounded-memory #9756

[oech3]

Is there any chance for creplacing LD_PRELOAD by strace one liner (at here or upstream)?

[ChrisDryden]

I'm not entirely sure for all cases but for some of them the actual libc calls are different so they wouldn't match exactly in strace too

[ChrisDryden]

Discovered that the nfs-removal-race.sh test is actually only working for GNU for glibc versions < 2.33 so it would have been virtually impossible to have used the same internal calls to get it to pass. I sent in a PR to update the test for GNU themselves and it should let us pass the test. I'm guessing some of the other LD_PRELOADS could pass if we update the tests to a newer glibc version

[ChrisDryden]

Given that more than half of the entries are closed here I was thinking of starting a new one to be able to reduce the noise and add more visibility into the strategies for some of the more difficult ones. Was also thinking that it would be good to do a branch run with the latest GNU tests to see which tests will begin failing with the latest integration tests updates. I've been following the GNU commits daily and have been trying to make PR's whenever new stuff gets tested but it would be good to have a tracker for new regressions so when we have a new version update we don't have a dip on the GNU compatibility graph again.

[sylvestre]

Sounds good for closing

We can create a new job to run on main in parallel of the release.

[oech3]

it would be good to do a branch run with the latest GNU tests to see which tests will begin failing with the latest integration tests updates.

How about just bumping the commit instead of running both?